A point about these election polls...

[mondialito]

Apologies if this has already been mentioned, but one thing that we shouldn't forget about these General Election polls is that this is the first election since the methodology for polling largely changed after the clusterf**k that was 2015. Adjusting sampling to ensure a greater proportion of those more likely to vote are captured has led to stronger support for the Tories whether behind or in front than in the run up to 2015.

However, couldn't there be a possibility that the new methodology has led to general overstatement of Conservative support? Not to extent that the Tories aren't really leading of course (this isn't The Canary), but that the lead isn't crushing landslide territory as others suggest? There is much evidence around that a landslide is indeed on the cards such as the local election results, feelings on the doorstep and where senior Tories turn up during the campaign (this could be more telling than it first appears, how we all laughed when Cameron turned up in Twickenham), but what if the clock strikes 10pm on 8th June and accurately projects a Conservative majority of 'only' 70? Could this be another case of pollsters screwing up again leading to another inquiry, this time asking why the polls projected a majority double what it really turned out to be?

[Adrian]

Can I get away with just replying "No"?

I'm getting the sense that the polls are in the right ballpark, and the local elections provided evidence that the swing to the Tories is quite sharp. There's quite a bit of variation though, so I don't think there's the clear message from them that you suggest. The range has been anything from a majority of 60 to 160. Maybe they'll start to settle down, one way or the other.

[johnr]

I think a lot of that variation can be explained with how the polls treat "Dont Knows". Some re-allocate back to their original parties a % of these. I suspect thats hitting us, and explains the polls with Labour in the mid 20s. The polls with higher Labour figures (30/31) is re-allocating back to Labour people who are saying "Dont Know" because they're not sure of COrbyn.

[Harry Hayfield]

I don't trust the polls (as I am a child of the 1992 debacle) so therefore what I do is every time a poll is published I look at the historical error this far from the election and then adjust it. For instance today's poll for Good Morning Britain said Con 47%, Lab 30, Lib Dem 7%. On average (since Election 1997) thirty days out from an election, the Conservative vote has been 1% higher on Election Night than 30 days out, Labour's vote has been 5% lower and the Liberal Democrats have been 2% higher (so therefore I believe the real levels are Con 48%, Lab 25%, Lib Dem 9%)

[yellowperil]

May 9, 2017 6:43:10 GMT @jamesdowden said:

I'm more sceptical about the polls because I doubt that the swing will be remotely uniform. Huge numbers of 2015 Ukip voters in Kent, Essex, Suffolk, Norfolk, and Lincolnshire voting Conservative this time will most probably result in a gain of one or two seats, because most of them are blue already.

But I wouldn't view a majority of 70 as a disappointment. There was a lot of comparing the slim majority gained at the 2015 General Election with previous elections with similar results, the best comparison being 1951. Therefore the target really has to be equalling or bettering Sir Anthony Eden's majority of 60 in 1955. So 70 would be decidedly on the good side.

and no doubt Theresa will make as good use of it as did Sir Anthony...

[The Bishop]

May 9, 2017 7:09:49 GMT Harry Hayfield said:

I don't trust the polls (as I am a child of the 1992 debacle) so therefore what I do is every time a poll is published I look at the historical error this far from the election and then adjust it. For instance today's poll for Good Morning Britain said Con 47%, Lab 30, Lib Dem 7%. On average (since Election 1997) thirty days out from an election, the Conservative vote has been 1% higher on Election Night than 30 days out, Labour's vote has been 5% lower and the Liberal Democrats have been 2% higher (so therefore I believe the real levels are Con 48%, Lab 25%, Lib Dem 9%)

But if pollsters are more pro-Tory in their methodological assumptions this time, "historic patterns" arguably aren't that relevant.....

[carlton43]

The problem with the polls is people!

1) Polls are a collation of answers to questions on a given day. They are not a forecast of anything.
2) Many people are Don't Know, Won't Say or are Perverse.
3) I have been polled on a number of occasions and usually say Labour to keep the Right on it's toes.
4) The sample is always too small and never representative.
5) No 'Methodology' can put that right for obvious reasons.
6) Having an appropriate 'Method' is ridiculous nonsense. It suggests one has perfect knowledge of the sample taken.
7) People have poor memories and forget how they voted last time.
8) People change their minds from day to day.

I like polls because despite their faults they are full of useful information. What we need is very careful sample preparation, no methodology, no exclusion of Don't Knows, no tinkering and full disclosure of the sample taken. Large samples and frequent samples. And some constant samples with repeat and repeat polling.

The trend is all. We can use the base material and run our own methodology against it.

[Adrian]

May 9, 2017 12:14:43 GMT carlton43 said:

The problem with the polls is people!

1) Polls are a collation of answers to questions on a given day. They are not a forecast of anything.
2) Many people are Don't Know, Won't Say or are Perverse.
3) I have been polled on a number of occasions and usually say Labour to keep the Right on it's toes.
4) The sample is always too small and never representative.
5) No 'Methodology' can put that right for obvious reasons.
6) Having an appropriate 'Method' is ridiculous nonsense. It suggests one has perfect knowledge of the sample taken.
7) People have poor memories and forget how they voted last time.
8) People change their minds from day to day.

I like polls because despite their faults they are full of useful information. What we need is very careful sample preparation, no methodology, no exclusion of Don't Knows, no tinkering and full disclosure of the sample taken. Large samples and frequent samples. And some constant samples with repeat and repeat polling.

The trend is all. We can use the base material and run our own methodology against it.

Considering how accurate polling is - and by and large it is pretty accurate, and I'm not just talking about political polling - some of what you say above is patently false. And if you want to see much more detailed information about each political poll, it's there on the pollsters' websites, including the raw unadjusted figures. As far as constant samples are concerned, the BES has done that at several elections (and in between) and obviously there's useful info come out of it about how people changed (and didn't change) their minds during campaigns, etc. Such polls are very dangerous as news items though because the headline figures in them tend to be more unreliable than in the one-off polls.

[Deleted]

I'm not one for taking things for granted, but if there's one thing that gives me confidence and indicates the opinion polls are broadly on the right track, it is the Copeland by-election result from just two and a half months ago which showed a 6.7% swing in what is a tight Lab-Con marginal. If anything, it pointed towards a Conservative lead slightly greater than the national polls were indicating at the time. I realise it's not the perfect indicator, but it's probably the best we have.

[finsobruce]

May 9, 2017 16:08:57 GMT swanarcadian said:

I'm not one for taking things for granted, but if there's one thing that gives me confidence and indicates the opinion polls are broadly on the right track, it is the Copeland by-election result from just two and a half months ago which showed a 6.7% swing in what is a tight Lab-Con marginal. If anything, it slightly pointed towards a Conservative lead slightly greater than the national polls were indicating at the time. I realise it's not the perfect indicator, but it's probably the best we have.

it's a terrible cliche that campaigns change nothing, but I get the feeling that this time the campaigns really will change nothing.... In scotland the local elections have provided some oomph and in England the collapse of UKIP will alter the results markedly in some places. So yes, I think Copeland might well be an accurate pointer to how things will go.

[marksenior]

I suggest you all read Martin Boon's comments on the latest ICM poll . He is saying he is pretty sure they have got Labour support in the right ball park but they are overstating Conservative support and understating Lib Dem support .
Amazingly he and ICM have seriously considered more methodological changes to correct this but decided it is not good to do so mid way in an elections campaign .
If the polls prove wrong , his excuse will presumably be , that we said we thought we were wrong but for consistency we could not correct the errors we knew were there .
That is not to say that the Conservatives will not win but betting on the spreads based on the polls is likely to prove costly .

[pepperminttea]

May 9, 2017 16:29:36 GMT marksenior said:

I suggest you all read Martin Boon's comments on the latest ICM poll . He is saying he is pretty sure they have got Labour support in the right ball park but they are overstating Conservative support and understating Lib Dem support .
Amazingly he and ICM have seriously considered more methodological changes to correct this but decided it is not good to do so mid way in an elections campaign .
If the polls prove wrong , his excuse will presumably be , that we said we thought we were wrong but for consistency we could not correct the errors we knew were there .
That is not to say that the Conservatives will not win but betting on the spreads based on the polls is likely to prove costly .

That's not quite what he said at all. He said and I quote: ''Clearly, if we are to take the PNS as the best evidence available of the current state of play, we’re over-stating the Tories and seriously under-representing the Liberal Democrats'' but as we know projected national vote share from the local elections is rubbish at predicting general elections especially because the Tories tend to underperform in local elections while the Lib Dems overperform. He later goes on to say ''That said, my view prior to 2015 was that we were over-stating the extent of their[Lib Dems] fall, but in the event we were largely not'' which perhaps points to the methodology being in the right ball park as far as the Lib Dems go this time.

The Lib Dems were at 18% in the local election PNS and I don't know of a single person who thinks that's the level they're at (or anywhere close to it) as far as the general election goes at the present time. It's possible that the Lib Dems are being understated by 1 or 2 points maximum nationally but PNS is an incredibly poor reason why.

A good reason for up weighting the Tories, Labour and the Lib Dems would be that UKIP is not going to run in very many constituencies so their national vote share is obviously inflated.

[No Offence Alan]

I think what has bothered the pollsters in recent years is increasing turnout.
For 2001 and 2005 it was relatively easy to use previous voting intention as a model and then allow for differential turnout and switchers.
But if turnout is increasing how do you model people who didn't vote last time around?

[marksenior]

May 9, 2017 17:10:46 GMT pepperminttea said:

May 9, 2017 16:29:36 GMT marksenior said:

I suggest you all read Martin Boon's comments on the latest ICM poll . He is saying he is pretty sure they have got Labour support in the right ball park but they are overstating Conservative support and understating Lib Dem support .
Amazingly he and ICM have seriously considered more methodological changes to correct this but decided it is not good to do so mid way in an elections campaign .
If the polls prove wrong , his excuse will presumably be , that we said we thought we were wrong but for consistency we could not correct the errors we knew were there .
That is not to say that the Conservatives will not win but betting on the spreads based on the polls is likely to prove costly .

That's not quite what he said at all. He said and I quote: ''Clearly, if we are to take the PNS as the best evidence available of the current state of play, we’re over-stating the Tories and seriously under-representing the Liberal Democrats'' but as we know projected national vote share from the local elections is rubbish at predicting general elections especially because the Tories tend to underperform in local elections while the Lib Dems overperform. He later goes on to say ''That said, my view prior to 2015 was that we were over-stating the extent of their[Lib Dems] fall, but in the event we were largely not'' which perhaps points to the methodology being in the right ball park as far as the Lib Dems go this time.

The Lib Dems were at 18% in the local election PNS and I don't know of a single person who thinks that's the level they're at (or anywhere close to it) as far as the general election goes at the present time. It's possible that the Lib Dems are being understated by 1 or 2 points maximum nationally but PNS is an incredibly poor reason why.

A good reason for up weighting the Tories, Labour and the Lib Dems would be that UKIP is not going to run in very many constituencies so their national vote share is obviously inflated.

Every recent ICM poll has found an increasing number of LD respondents and the more they find they more they weight them down . The last poll had 12% Lib Dems in the sample weighted down to a final 9% . ICM are doing this purely because they overstated the Lib Dems at the last election . In my opinion the circumstances now are not the same and the changes are an over correction . The Lib Dems are unlikely to match the 17% NEV in the local elections ( though there have been occasions when they have done so in the past ) but equally the NEV has never exceeded the GE performance by more than 3 to 4 % .
but in no case

[John Chanin]

May 9, 2017 16:29:36 GMT marksenior said:

I suggest you all read Martin Boon's comments on the latest ICM poll . He is saying he is pretty sure they have got Labour support in the right ball park but they are overstating Conservative support and understating Lib Dem support .
Amazingly he and ICM have seriously considered more methodological changes to correct this but decided it is not good to do so mid way in an elections campaign .
If the polls prove wrong , his excuse will presumably be , that we said we thought we were wrong but for consistency we could not correct the errors we knew were there .
That is not to say that the Conservatives will not win but betting on the spreads based on the polls is likely to prove costly .

This is certainly possible, as debate about the changes to polling adjustments has shown.  There have been adjustments relating to previous vote (and recall of previous vote) where electors are assumed in a certain percentage to revert to their previous vote, and those saying don't know are allocated in certain proportions to the party they voted for last time.  But the Liberals were crushed at the last election.  So the polling companies are downgrading the proportion of electors who now say they will vote Liberal.  This may prove to be incorrect, but we won't know until the election happens. 

The polling companies in a desperate effort to get the results right, as their commercial credibilty depends on it, are now making large adjustments to the raw data they collect, well beyond simply ensuring demographic consistency as they used to do.  And although the raw data is published for every credible polling company, the exact nature of the adjustments they make to it is not.  Nonetheless I suggest anyone interested in polling takes a look at the raw data.  

[brianj]

There's now another issue that must be considered in polling from here on in: the Greens will not have candidates in constituencies covering a quarter of Great Britain, and UKIP won't have candidates in constituencies covering 40% of Great Britain. Both parties will therefore underperform their polling averages, to the benefits of other parties (primarily the LibDems and Tories, respectively).

[The Bishop]

May 12, 2017 21:49:23 GMT brianj said:

There's now another issue that must be considered in polling from here on in: the Greens will not have candidates in constituencies covering a quarter of Great Britain, and UKIP won't have candidates in constituencies covering 40% of Great Britain. Both parties will therefore underperform their polling averages, to the benefits of other parties (primarily the LibDems and Tories, respectively).

I know the Green party leadership is if anything closer to the LibDems than Labour these days, but I am not convinced the same is true of most of their voters.

[greenchristian]

As for where the Green & UKIP vote goes in constituencies where we/they aren't standing a candidate, it's worth remembering that both parties attract anti-establishment votes. Those voters are less likely to go Conservative/Labour/Liberal Democrat in the event that their usual anti-establishment party isn't standing.

[thirdchill]

May 13, 2017 20:42:12 GMT greenchristian said:

As for where the Green & UKIP vote goes in constituencies where we/they aren't standing a candidate, it's worth remembering that both parties attract anti-establishment votes. Those voters are less likely to go Conservative/Labour/Liberal Democrat in the event that their usual anti-establishment party isn't standing.

Agreed. There is a surprising amount of both who second preference each other when this system is in place, particularly if the only other choices are Lab/Con/LD.

Can imagine a fair few staying at home tbh. And not going for the 'establishment' party they are supposed to be voting for.

[Sibboleth]

I know of one former LibDem supporter who at the last GE was undecided between UKIP and the Greens (not sure how they went in the event). People like this aren't particularly easy to model for, which is an issue for the polling industry these days. Except it sort of isn't as they by definition are not voters whose votes decide outcomes.